COBRA is a computer-designed bridge bidding system, developed by E. T. Lindelof. The system is precisely defined, which makes it possible for computers to use it. Given the current quality of computer bridge-playing programs, it would be interesting to see if a computer can actually bid on an expert level. My analysis gives an unclear result; COBRA lost about half an IMP per board to the humans in the 1987 World Championships, but I don't have a perfect understanding of the system. The system is similar to the Schenken Club, as described in Schenken's _Better Bidding in 15 Minutes_. It does include some additional refinements and conventions. Lindelof published a book which describes the system; it is available from IPBM 455 Alfreton Road Nottingham NG7 5LX England Phone: +44 1159 422615. The book doesn't give a complete description of the system, but the description is good enough to see how COBRA thinks and bids. Lindelof tested COBRA on bidding challenge hands from Bridge Magazine, and the hands from the 1959 Bermuda Bowl. The Bermuda Bowl results, in 96 boards: Italy vs. COBRA: COBRA improved the North American score by 98 IMPS. North America vs. COBRA: COBRA improved the Italian score by 69 IMPS. Italy actually won 214-189; thus, if these results are correct, either side could have won by a large margin if it had used COBRA as its bidding system. The gain of nearly one IMP per board seemed unusually high, and I was suspicious of COBRA's results on a few hands, so I decided to check the results for myself. I have a book of the 1987 Bermuda Bowl and Venice Trophy, with all interesting hands from the final detailed, so I used that for the sample. My criteria were similar to Lindelof's. I didn't make a comparison if I couldn't tell how the human bidding would continue, or on a few hands in which the description of COBRA in the book was inadequate. Also, some of the British pairs played a forcing pass system, against which I often had no idea how COBRA would defend. If the humans at the same table reached the same denomination (but not necessarily the same level), COBRA was usually credited with the same number of tricks. I made an allowance if the auction with COBRA would have caused a different result. For example, there was one hand on which COBRA allowed the opponents to make a overcall, which directed the best opening lead; it was credited with the tricks made against this lead, rather than the tricks taken by human declarers whose auction made the overcall impossible. If humans at other tables reached the same denomination, I tried to pick the normal result from among those results the humans reached. Again, allowances were made if necessary. If COBRA reached a denomination not reached in any room, I worked out a reasonable result on my own, after reading the expert commentary. I have checked the first 48 boards carefully so far. COBRA's total swings: -10 for US men, -20 for Great Britain, -21 for France, -61 for US women, total -112. This is much worse than Lindelof's result. Part of the reason may be that I don't understand COBRA fully (using the book rather than the full program). I don't know whethe COBRA would do if used perfectly. I looked at all the significant swings (5 IMPs or more), and broke them down by cause, to see where COBRA gained its advantages. A swing was attributed to luck as well as to another cause if it was not the result of reaching a better contract. For example, on the sample hand, COBRA reached a good slam which none of the humans reached, but it went down. COBRA lost 18 IMPs, but this was not a result of COBRA's bad judgment, so it was scored as judgment luck. System swings (fundamental differences in the system led to a different contract being reached, for reasons unrelated to judgment): +64 and -64 on eleven boards. COBRA's judgment (decisions to bid slam, play 5C rather than 3NT, etc., which were made by COBRA but not by most of the humans): +11 and -143 on ten boards. The expert consensus seems to be better. COBRA's judgment luck: +44 and -18 on three boards. COBRA went down in a good slam (-18, because two of four humans went down in games on the same bad breaks). avoided a good game which was doubled when trumps didn't break (+16), and made a bad slam (+26). Humans' judgment (COBRA agreed with the human consensus, but one human pair bid differently, or two human pairs had different results, neither of them COBRA's): +47 and -22 on seven boards. Humans' judgment luck: -25 on two boards. France bid and made a mediocre slam (-12) and the US women bid and made a bad slam (-13). Human misunderstandings: +11 on one board. The US women had a misunderstanding whether 4NT was Blackwood or natural, and got to a slam off two cashing aces (+11). Thus it seems that COBRA's judgment may be slightly better than the judgment of any individual expert pair, and comparable to the consensus of experts. I don't know whether humans playing COBRA could do as well; I had to keep referring back to the book to check point ranges and defensive methods. In fact, in an earlier version of this discussion, I posted two examples of COBRA auctions which reached slams. I then rechecked the book and discovered that the second example auction, which reached a mediocre slam, was incorrect; removing this example from the totals above gained an extra 29 IMPs for COBRA.